We used two types of UCNPs of the structure NaYF4:Yb: Er and NaYF4:Yb:Er@NaYF4, termed E1- and E2-UCNP, respectively. The synthesis of NaYF4:Yb, Er@NaYF4 nanoparticles was based on the Ostwald ripening-mediated method with some modification [17]. First, cubic NaYF4 nanocrystals were prepared by mixing 1 mmol YCl3, 6 mL oleic acid and 10 mL of octadecene in a 50-mL flask. The resulting mixture was heated to 150 °C under airflow under constant stirring for 30 min to form a light-yellow solution that was cooled down to room temperature. Then 10 mL of methanol solution with 1.6 mmol of NH4F and 1 mmol of NaOH was added and stirred for 30 min. After heating to remove methanol, the solution was cooled. The solution was heated to 290 °C under argon flow using vigorous stirring for 45 min and cooled to room temperature. NaYF4 nanoparticles obtained were collected and resuspended in a solution of 5-mL oleic acid, 8-mL octadecene. 0.2 mmol core UCNPs in 15-mL cyclohexane were added to a 100-mL flask, mixed with 10-mL oleic acid and 16 mL octadecene. The resultant core-shell NaYF4:Yb, Er@NaYF4 nanoparticles were collected and washed with centrifugation and dispersed in cyclohexane. Note that E1-UCNP is characterised by a much shorter emission lifetime in comparison with that of E2-UCNP.

Another type of UCNP, β-NaYF4:Yb3+, Tm3+/NaYF4, here termed T-UCNP, were synthesised using a procedure similar to that described above and reported in detail in [18].

The conversion coefficient of NaYF4:Yb:Er @ NaYF4 (ηUC) was evaluated to be about 2 %. To improve it, we synthesised core-shell UCNP of the structure β-NaYF4:Yb3+, Tm3+/NaYF4 termed T-UCNP, as reported in detail in [17]. The conversion coefficient of T-UCNP was measured to be as high as 11 % (Fig. 3B) and is much greater than ordinary two-photon absorption [13]. The distinct spectral bands of these particles are shown in the Fig. 1C.

The conversation efficiency of UCNPs is presented in Fig. 3B [17].

All experiments with animals were performed using general anaesthesia (Zoletil mixture, 40 mg/kg of the animal's weight and 2 % Rometar, 10 μL, 10 mg/kg) using intraperitoneal injection following the “Guide for the Care and Use of Laboratory Animals”. The research was approved by the Ethics Committee of N.I. Lobachevsky State University of Nizhny Novgorod from 03.03.2021, protocol № 50. Two mice were housed in a single cage on a 12-h light/dark cycle under normal lighting conditions in transparent polycarbonate cages with an international standard stainless steel grid with ad libitum access to food and water using full-fat extruded feed for laboratory animals “Chara” and filtered (“Aquafor”, Russia) water with complete replacement of water in drinking bottles 2 times a week.

Professional autoclavable material REHOFIX MK 2000 (granules from the core part of corn cobs) was used as bedding material, which was replaced every two days during the preparation and conduction of the experiment. Cleaning of the cages (washing the plastic base) was also performed once every two days. The air temperature was maintained within the range of 20–24 °C, with a relative humidity of 50 ± 20. To administer two types of UCNPs in live anaesthetised animals, we used 2 males at 8 weeks of age BALB/c mice weighing 20–25 g to examine the biodistribution of these particles after a single intravenous injection. The fur around the chest, abdomen and back areas of the animals was shaved to reduce scattering of the PL signal. 100 µL 0.05 % wt. PA-UCNPs/PA colloidal solution (1 mg/kg) of phosphate buffer solutions, pH 7.2 was injected in a lateral tail vein. The detection of the PL was carried out on either the ventral or dorsal sides of the animals. The animals were euthanized 2 h after PA-UCNPs/PA injection by cervical spine dislocation. The liver was extracted and used for the imaging experiments.

To demonstrate the advantages and limitations of the proposed lifetime imaging, we developed a simplified theoretical approach [15].

Consider a one-dimensional case of the laser-scanning PL imaging configuration, as shown in Fig. 2. A PL signal at the photomultiplier S(t) is modelled using the following expression:

(1)

G represents the intensity distribution of the excitation laser at 975 nm. G depends on the space and time variables, x, t, respectively, which are functionally dependent as x–vt, and describe a travelling wave propagating in the positive x-direction with the speed v. P(x) represents the UCNP sample profile. The UCNP PL versus time is described by Г(t) [19], with the time trajectory shown in Fig. 1B. This time trajectory features the PL signal rise and is explained by the complex upconversion energy transfer processes. The second phase of decay is determined primarily by the UCNP PL lifetime τph. Assume , and a Gaussian function intensity distribution , with the beam waist, w equal to the focussed beam diameter.

Consider the sample as a discrete set of UCNPs with diameters << w. A single UCNP is modelled as a delta-function positioned at = 0, i.e. P(x)=δ(0). If τph is comparable with the scanning time, i.e. τph~ τs the expression (1) is transformed to:

(2)

where erf – the error function, where (xi,yi) – T-UCNP coordinates, Ai – the relative amplitude of the signal. Evaluation of the expression (2) leads to the conclusion that this is a convolution of the surface profile of UCNP with the time-decay function of UCNP, i.e. temporary function Г(t). One can think of S as the impulse function, which can be used for deconvolution, where two parameters need to be adjusted for the best results: τph and w. PL signal can be acquired by rapidly scanning through the sample, and the surface distribution of UCNP is reconstructed at the post-processing stage. It was straightforward to generalise the 1D-case to 2D-case.

The theoretical modelling of single-particle T-UCNP was carried out using the following formula, where vt is replaced by the time-dependent coordinate x:

(3)

Where (xi,yi) is T-UCNP coordinates, Ai – is the relative amplitude of the signal.

To restore an undistorted T-UCNP image, we used the Lucy – Richardson algorithm. Fig. 4 (right panel) represents an image processed using the Lucy – Richardson algorithm, in which expression (3) was used as the point spread function (PSF) input function under the summation sign.

The result of the numerical simulation of a 2D PL image of a set of UCNPs randomly distributed in the field of view is shown in Fig. 4A randomly distributed noise was also added to test the robustness of the image reconstructions procedure. Note the significant reduction of the original image size. We emphasis a significant improvement of the signal-to-noise ratio evident by visual examination of the salt-and-pepper noise (Fig. 4A). This noise is unobservable in the reconstructed image (Fig. 4B).

Rapid image acquisition was realised by setting the excitation intensity to a value higher than the saturation power density of E2-UCNPs (>0.1 Jcm-2) easily achievable at the typical settings of the laser-scanning microscope. The focussed laser beam was scanned at a speed of 10 cm/s and dwelled about 10 μs at each pixel. The specimen represented a thin slice of the liver tissue extracted from a mouse 3-h post-injection of E2-UCNPs prepared in phosphate buffer saline. Fig. 5A shows an overlay of the bright-field image of the liver tissue and PL image of E2-UCNPs accumulated in this tissue. Comet-like PL signals from single or clustered UCNPs are readily observable. Fig. 5B presents these signals as a surface plot. The image deconvolution procedure was applied, and the result is presented in Fig. 5C. The blurred PL image was converted into a set of intense narrow discrete peaks of about 1.5-μm in diameter comparable to the diffraction-limited spots. The E2-UCNP PL lifetime was determined to be 0.8 ms in good agreement with the expected value (Fig. 1B).

PL lifetime of UCNPs (τph) represents an important configuration parameter for the deconvolution procedure. We measured τph using an alternative method employing a setup equipped with an optical modulator (chopper) and time-correlated single-photon counting module (TCSPC) by processing images of the sample scan. For each particle shown in Fig. 3C, PL attenuation curves in the spectral range of 446–486 nm were recorded using the scheme shown in Fig. 3A. The obtained curves were approximated by a dependence of the form

(4)

with the lifetimes of the exponential components τ, and the amplitudes of the exponential components a.

The parameters τm and ti were calculated to determine the average lifetime of the studied T-UCNP:

τm – the average lifetime of the components of a multi-exponential decay weighted by their amplitude coefficients (a1,a2 etc.). For a two-exponential decay, τm is defined as expression (5):

(5)

ti – the average lifetime of the decreasing components weighted by their integral intensities. The intensity-weighted value of ti for a two-exponential decay is defined as expression (6):

(6)

The mean lifetime ti is more sensitive to changes in the slow component of the lifetime, while the mean lifetime τm is more sensitive to changes in the fast component.

Then, without changing the position of the sample, the sample was scanned at the scanning speed corresponding to the pixel dwell value of 12.61 µs. The resulting image is shown in Fig. 3D.

The PL signal profile of each particle was extracted for subsequent deconvolution using the view function [expression (6)]:

(7)

where I is the intensity of the nanoparticle's PL signal; t0 and t are the sample scanning time (t0) is the time corresponding to the moment of the beginning of PL signal growth when the scanner passes through the sample area in which the particle is located); since the PL signal profile obtained from the microscope is a dependence of the PL signal on the scanner position (in pixels), to convert the abscissa scale from x (pixels) to t(µs), the initial values of x were multiplied by the scanning time of one pixel (Pixel Dwell); τ1 – parameter corresponding to the UCNP PL signal decline time; τ2 – parameter corresponding to the T-UCNP PL signal rise time; A and B – amplitude factors; C – the background signal value (noise level).

τph of T-UCNP measured by the multi-photon microscopy (MPM) method, as well as τm and ti values obtained using TCSPC, are shown in Fig. 6.

One can see that the deconvolution method provides an underestimate of τUC in comparison with the more reliable method based on the chopper/TCSPC module.

To demonstrate the capability of multiplexing/demultiplexing, we carried out lifetime imaging using 2 types of UCNPs embedded in excised animal tissue. As the long and short emission lifetime UCNPs we selected T- and E1-UCNPs, respectively. A cocktail of UCNPs was administered in a live laboratory animal, and the extracted liver was sliced and examined (see M&M for details). Fig. 7A shows a photograph of a mouse liver containing two types of particles, obtained using a raster scan with a long dwell time in the laser beam per pixel (177 μs). By measuring the PL spectra of each pixel as shown in Fig. 7C by measurements at points (1) and (2) of Fig. 7A, we were able to identify T- and E1-UCNPs, since the PL spectrum of the T-UCNP particles was distinguished by pronounced peaks in the 450–508 nm (blue) region, whereas E1-UCNP featured characteristic spectral bands in green and red.

Next, the same sample was scanned using the MPM system (Fig. 7B). In the region where we were able to identify T-UCNP from the previous experiment, we observed a blurring of the point in a line along the scan-axis, which was absent in the region where the short-lived E1-UCNP was located. This was due to the much longer lifetime of T-UCNP compared to E1-UCNP. Thus, it became possible to distinguish, identify particles in the sample by the character of the observed comet-like blurring of the point. Besides, τUC can be evaluated.

We demonstrated a method to rapidly acquire images of discrete photoluminescent nanomaterials by operating a commercial nonlinear optical microscopy in a non-descanned detection mode. The resultant blurred patterns produced by discrete PL nanoparticles such as UCNPs were processed by a deconvolution algorithm to achieve an almost diffraction-limited PSF. Besides, we demonstrated the reconstruction of τph. Since it is possible to engineer τph of UCNP, we produced 2 types of UCNP with long and short τph to demonstrate multiplexing and demultiplexing to identify both types of UCNPs. This capacity was useful to discriminate different UCNPs in turbid biological tissue [16].

Among the shortcomings of the reported method, the accuracy of determination of τUC is limited by the signal-to-noise ratio – the method tends to underestimate τph, as it was the case shown in Fig. 3D. Secondly, the increased excitation intensity leads to the broadening of PSF of UCNPs and can compromise the resolution. For example, Fig. 7 shows discrete nanoparticles sized several micrometres in diameter. This limitation can be circumvented by using unique types of T-UCNP [20], but this limits the repertoire of available PL nanomaterials. Thirdly, the multiplexing capacity is limited by the image size and laser-scanning rate, which can accommodate the longest τph, while the shortest τph may remain undetectable. This scenario is demonstrated in Fig. 7B, where E1-UCNP featuring short τph showed negligible streaking, while T-UCNP showed the streaking of the length comparable with the image size.

In this communication, we demonstrated a new approach for rapid PL (phosphorescence) image acquisition of samples containing PL materials, such as biological specimens labelled with discrete UCNPs using laser scanning multiphoton microscopy. The comet-like blurred images of discrete UCNPs obtained as a result of the rapid scanning were reconstructed by applying the deconvolution procedure. The contrast of discrete UCNP images was enhanced, and the UCNP emission lifetimes were determined. This enabled demonstration of multiplexing/demultiplexing using lifetime imaging mode, where the UCNP lifetime was engineered. The power of this method was demonstrated by the identification of 2 types of UCNPs accumulated in the liver of a laboratory animal. We believe that the demonstrated method can be useful for rapid lifetime imaging where several molecular specific labelling agents are required.

Andrey V. Yudintsev carried out experiments, Artyom O. Zviagintcev and Alireza Maleki implemented image reconstruction algorithms, Vladimir A. Vodeneev developed the signal processing and image reconstruction algorithms, Andrei V. Zvyagin conceived the main concept, coordinated the project and took part in writing the manuscript together with Artyom O. Zviagintcev and Alireza Maleki. All authors participated in the discussion and editing of the work. All authors approved the final version of the publication.

Supplementary materials associated with this article can be found in the online version at doi: https://doi.org/10.47093/2218-7332.2022.338.06.S

Conflict of interests. The authors declare that there is no conflict of interest.

Financial support. The authors wish to acknowledge financial support from the Russian Foundation of Basic Research (RFBR), grant No 20-04-00182.